In this paper, we propose a new decentralized method for structural health monitoring based on substructure dynamical parameter identification, which can be executed in parallel on multiple smart sensor nodes. A decentralized Kalman filter algorithm is applied for modal parameter identification of structures using multi-input and multi-output ARX model. Furthermore, network communication topology is investigated using the substructure approach for finding a practical application of the decentralized algorithm. Numerical simulation studies are carefully performed for three types of network topologies and two types of model structures. Identification tests using the seismic observation data of existing 5-story building are also conducted. The results show the feasibility of the proposed decentralized method.